# Alphabet expands TPU edge in AI compute race > Source: > Published: 2026-06-27 13:21:04+00:00 ### What happened CNBC reports that Alphabet's in-house Tensor Processing Units (TPUs) serve as the compute engine for its Gemini chatbot and as a core part of Google Cloud's AI compute offering. Customers including Anthropic rent access to TPUs on Google Cloud, and in some cases customers can now buy TPUs for their own data centers. CNBC cites FactSet projecting Google Cloud revenue to rise roughly 64 percent this year to $96 billion, and says analysts model growth above 50 percent for 2027. Google Cloud reported over $20 billion in Q1 2026 revenue, up 63 percent year over year. CNBC quotes Brad Gastwirth saying Google is "probably the most underappreciated competitor of Nvidia." ### The Blackstone joint venture The Google-Blackstone joint venture, announced in an official Blackstone press release on May 18, 2026, commits $5 billion in initial equity from Blackstone-managed funds to bring 500 MW of TPU compute capacity online by 2027, with plans to scale over time. Google executive Benjamin Treynor Sloss, who spent over two decades building Google's global infrastructure and operations, was named CEO of the new company. The venture gives customers a compute-as-a-service path to TPUs outside of standard Google Cloud access. Blackstone President and COO Jon Gray described it as "a generational opportunity to invest capital at scale building AI infrastructure." Google Cloud CEO Thomas Kurian confirmed the JV "helps meet growing demand for TPUs, which are optimized specifically for efficiency and performance in the AI era." ### Technical context Custom accelerators optimize the matrix-multiply and memory-access patterns central to large-model training and inference. Google announced its eighth-generation TPUs at Google Cloud Next in April 2026, splitting the line into TPU v8t for training and TPU v8i for inference, signaling continued hardware investment. CNBC places TPUs as a differentiated piece of vertical integration: companies building custom silicon typically improve performance-per-dollar and energy efficiency versus general-purpose GPUs. For practitioners, this dynamic shapes cost modeling, workload placement, and procurement choices when selecting cloud versus on-prem compute. ### Context and significance Custom AI accelerators change the economics of large-scale model development by shifting more of the compute stack from third-party vendors to cloud providers or their captive silicon. The Blackstone venture adds a private capital structure that gives enterprise customers a third path to TPU access beyond Google Cloud subscriptions or direct on-prem purchase. The commercial pipeline - Anthropic, the Blackstone JV, and on-prem sales - turns Alphabet's decade of internal chip investment into an externally monetizable infrastructure product. ### What to watch Track independent benchmarks and performance-per-dollar comparisons between TPUs and GPU alternatives, how Google Cloud pricing evolves as demand scales, and whether the Blackstone JV's 2027 500 MW target stays on schedule. Also monitor updated FactSet and analyst forecasts for Google Cloud revenue as Q2 2026 results approach, and watch for tooling or framework support that broadens TPU compatibility beyond early customers. ## Key Points - 1Alphabet's TPUs power Gemini and are rented or sold to cloud customers, creating a hardware-led revenue channel that broadens beyond Google's own products. - 2The $5 billion Blackstone joint venture targets 500 MW of TPU capacity by 2027, adding a private-capital structure for enterprise compute-as-a-service access. - 3Growing demand for rented and on-prem TPUs tracks Google Cloud revenue projections above 60 percent growth, altering procurement decisions for AI infrastructure buyers. ## Scoring Rationale The story covers a meaningful infrastructure shift: a confirmed $5 billion Google-Blackstone joint venture, confirmed on-prem TPU sales to enterprise customers, and strong Google Cloud revenue growth projections. Added authoritative sourcing (Blackstone official press release, Google Cloud Q1 data) strengthens the evidence base. Score is appropriate for a notable infrastructure development affecting procurement decisions and the Nvidia competitive landscape, though it is not a frontier-model or regulatory milestone. Practice with real Ad Tech data 90 SQL & Python problems · 15 industry datasets [Active Search Campaigns by BudgetEasy](/problems/sql/active-search-campaigns-by-budget) [High CPC Clicks & Poor Landing PagesMedium](/problems/sql/high-cpc-clicks-poor-landing-page) [Campaign ROAS by Attribution ModelHard](/problems/sql/campaign-roas-by-attribution-model) 250 free problems · No credit card [See all Ad Tech problems](/problems/datasets/adtech)